Automated Wall Detection in 2D CAD Drawings to Create Digital 3D Models

Generating digital 3D buildings models from scratch is time consuming and labor intensive. In this paper, we present an automated detection process leveraging computer vision and the information available in 2D drawings to reduce 3D modeling time. The recognition system is limited to walls and has two parts: (1) Image classification on walls by ResNet-50 model, (2) Object Detection on walls by YOLOv3 model. The system accepts new 2D drawings and outputs parameters of recognized walls. The parameters are input into Dynamo for 3D model reconstruction. We anticipate these types of systems, which rely on 2D drawings as recognition priors, will be pivotal to the industry's transition from 2D to 3D information modalities.